We build and test IMP on Windows, built with the Microsoft Visual Studio compilers (we use Visual Studio 2015 for both the 32-bit and 64-bit Windows .exe installers; for the Anaconda packages, which are built by conda-forge, VS 2019 is used). One complication is that different packages are compiled with different versions of Visual Studio, and mixing the different runtimes (msvc*.dll) can cause odd behavior; therefore, we recommend building most of the dependencies from source code using the same version of Visual Studio that you're going to use to build IMP.

We recommend building within the Anaconda environment, since many of the dependencies are already built, and the procedure is scripted so it is more easily reproducible:

• Install the right version of Microsoft Visual Studio (it is free, but registration with Microsoft is required). Current Anaconda policy is to build packages using the same version of Visual Studio that was used to build Python.
• Get and install Miniconda or the full Anaconda environment.
• Install necessary conda packages for development: conda install conda-build unxutils
• Then for each package foo you can start a Visual Studio command prompt, cd to the directory above the foo directory containing the conda build recipe then build it with conda build --python=3.9 foo
• Feel free to refer to our conda recipes for IMP and all needed dependencies. In particular, each recipe contains a meta.yaml file that applies any needed patches to make things work on Windows, and a bld.bat file that automates the build itself.

If you want to build outside of the Anaconda environment, the basic procedure we employed is as follows:

• Install Microsoft Visual Studio (it is free, but registration with Microsoft is required).
• Get and install cmake.
• Get Python (make sure you get the 32-bit version if you're going to build IMP for 32-bit Windows).
• Get and install the zlib package (both the "complete package, except sources" and the "sources" installers).
  • The package without sources can be installed anywhere; we chose the default location of C:\Program Files\GnuWin32. The sources, however, must be installed in a path that doesn't contain spaces (otherwise the Boost build will fail). We chose C:\zlib.
  • We found that the zconf.h header included with zlib erroneously includes unistd.h, which doesn't exist on Windows, so we commented out that line (in both packages).
• Download the Boost source code (we extracted it into C:\Program Files\boost_1_53_0), then
  • Open a Visual Studio Command Prompt, and cd into the directory where Boost was extracted
  • Run bootstrap.bat
  • You may need to help the compiler find the zlib header file with set INCLUDE=C:\Program Files\GnuWin32\include
  • Run bjam link=shared runtime-link=shared -sNO_ZLIB=0 -sZLIB_SOURCE=C:\zlib\1.2.3\zlib-1.2.3
• Get and install SWIG for Windows
• Get the HDF5 source code
  • Make a 'build' subdirectory, then run from a command prompt in that subdirectory something similar to cmake.exe -G "Visual Studio 10" -DHDF5_ENABLE_SZIP_SUPPORT:BOOL=OFF -DHDF5_ENABLE_Z_LIB_SUPPORT:BOOL=ON -DHDF5_BUILD_HL_LIB:BOOL=ON -DZLIB_INCLUDE_DIR="C:\Program Files\GnuWin32\include" -DZLIB_LIBRARY="C:\Program Files\GnuWin32\lib\zlib.lib" -DBUILD_SHARED_LIBS:BOOL=ON ..
  • Open the resulting HDF5 solution file in Visual Studio, change to Release configuration, then build the hdf5 project.
• (Optional) Build CGAL from source code.
• (Optional) Download the FFTW DLLs and follow the instructions at that website to make .lib import libraries needed for Visual Studio.
  • Copy libfftw3-3.lib to fftw3.lib and libfftw3-3.dll to fftw3.dll to help cmake find it
• (Optional) Get the GSL source code and build it:
  • Open the libgsl project file in the src\gsl\1.8\gsl-1.8\VC8 subdirectory
  • Build in Release-DLL configuration
  • Copy the generated libgsl.dll and libgslcblas.dll to a suitable location (we used C:\Program Files\gsl-1.8\lib)
  • Copy the corresponding .lib files, libgsl_dll.lib and libgslcblas_dll.lib (we recommend removing the _dll suffix and the lib prefix when you do this so that cmake has an easier time finding them, i.e. call them gsl.lib and gslcblas.lib).
• (Optional) Get numpy and scipy to match your Python version.
• (Optional) Get and install libTAU
  • Copy libTAU.lib to TAU.lib to help cmake find it.
• (Optional) Get the OpenCV source code and build it by following these instructions
  • Copy each opencv_*.lib to a similar file without the version extension (e.g. copy opencv_ml244.lib to opencv_ml.lib) to help cmake find it
• Set PATH, INCLUDE, and/or LIB environment variables so that the compiler can find all of the dependencies. (We wrote a little batch file.)

• Set up IMP by running something similar to

  cmake <imp_source_directory> -DCMAKE_BUILD_TYPE=Release -DCMAKE_CXX_FLAGS="/DBOOST_ALL_DYN_LINK /EHsc /D_HDF5USEDLL_ /DH5_BUILT_AS_DYNAMIC_LIB /DWIN32 /DGSL_DLL" -G "NMake Makefiles"

• Note: you may need to add /bigobj to CMAKE_CXX_CFLAGS, particularly if building for 64-bit Windows.
• Then use simply 'nmake' (instead of 'make', as on Linux or Mac) to build IMP. (cmake can also generate Visual Studio project files, but we recommend nmake.)
• To use IMP or run tests, first run the setup_environment.bat file to set up the environment so all the programs and Python modules can be found. (This batch file needs to be run only once, not for each test.)